COP26: A climate change read of artificial intelligence ethics and regulations

27 octobre 2021

Photo by Chris LeBoutillier on Unsplash

By: Eleonore Fournier-Tombs, Director, Inclusive Technology Lab

University of Ottawa Faculty of Law, Civil Law Section, Accountable AI in a Global Context

The 26th United Nations Climate Change Conference (COP26), will take place in Glasgow from October 31 to November 12, 2021. As in most sectors, artificial intelligence plays a growing role in the response to the climate crisis, notably for the prevention, mitigation and adaptation to climate change and the protection of vulnerable populations. A resource-intensive technology, it also can contribute to climate change and environmental degradation, if not implemented properly.  

In December 2021, UNESCO will publish its recommendation on the ethics of artificial intelligence, which aims to set standards for the regulation of AI in United Nations member countries. What are the links between international standards in climate change and in AI ethics, and how can COP26 further inform the regulation of AI systems?

The objective of the UNESCO recommendation is to provide a normative framework for United Nations member countries as they set about regulating AI. The recommendation contains universal values – which AI technologies should promote rather than endanger. It also contains principles, which can be seen as AI systems design guidance.

Environmental considerations are included in both the values and the principles sections. In addition, it discusses vulnerable groups, which implicitly describes those that will be worst impacted by a lack of action in the climate crisis, notably migrants, refugees, and those living in poverty.

Embedding environmental values into national regulations

The unregulated development of AI technologies is known to have adverse impacts on ecosystems that are felt throughout the technology’s life cycle.

UNESCO makes the following recommendations when it comes to ecosystem-friendly design:

  • Reducing the carbon footprint
  • Considering environmental risk factors
  • Eliminating unsustainable exploitation of natural resources

Researchers have estimated that the training of one big AI language model gives off as much carbon emissions as 125 roundtrips between New York and Beijing, or 300,000 kilos. Big technology companies in general have been criticized for their lack of consideration of their carbon footprint. The Tech Workers Coalition have argued that these companies have ongoing contracts with fossil fuel companies, have funded climate change deniers, and repressed climate refugees.

Similarly, AI technologies require natural resources that are often extracted in ways that not only damage the environment but also harm the human rights of local populations. For example, the mining of cobalt in the Democratic Republic of the Congo is known to be both unsustainable and damaging to human rights. Several big tech firms, including Apple, Google, Dell, Microsoft, and Tesla, were named in a case by the families of children injured or killed while mining cobalt. The mines also have a negative environmental impact, including radioactive waste, groundwater pollution and deforestation. Cobalt, however, is used not only in smartphone batteries but also in electric cars. This means that, as the world turns towards sustainable energy sources to fight climate change, mining practices will urgently need to be reviewed.

Figure 1: Climate and environmental concerns along the lifecycle of AI systems
Figure 1: Climate and environmental concerns along the lifecycle of AI systems

These are only a few of the climate and environmental risks of AI technologies. The technologies, however, can be built both sustainably and at the service of sustainability. The UNESCO document therefore recommends principles that should be integrated into the design of AI systems.

Requiring technologies that are sustainable

Various AI technologies have been documented to either support the achievement of or threaten the attainment of the Sustainable Development Goals (SDGs), depending on their use. Three SDGs in particular address the themes of COP26 – climate action, life below water and life on land (13, 14, and 15). Two other SDGs address a sustainable economy that would improve the capacity to reach climate objectives – affordable and clean energy, and sustainable cities and communities (7 and 11).

Researchers at the Linné FLOW Center in Sweden found that AI technologies more often had positive effects on the SDGs than detrimental ones. There have been numerous AI innovations for the climate, including in developing circular economies and reducing the use of natural resources. However, as we have seen, just because an AI technology has a sustainable objective does not mean that it is developed ethically.

Figure 2: Considering sustainability in purpose and design of AI systems
Figure 2: Considering sustainability in purpose and design of AI systems

Using AI with those affected by climate change

Climate change increases the incidence of both sudden-onset (storms and floods) and chronic (droughts) disasters, which often leads to forced migration. In its 2007-2008 Human Development Report, UNDP detailed different climate scenarios, according to which hundreds of millions of people could be displaced yearly by climate change. In 2020, IDMC reported 55 million displaced within their countries due to extreme weather events, and that number is set to increase.

AI technologies are used more and more in climate migration contexts, including in:

  • Predicting the probability of severe weather events and protecting the people affected by it, to prevent forced migration;
  • Predicting the number of people that will have to move and were, in order to better prepare humanitarian aid;
  • Organising visas and asylum (through biometric identification and automated case processing systems) for those that are permanently displaced.

However useful these technologies can be, they are also risky, as they impact populations living in highly precarious situations, where climate crises can be layered with conflict, discrimination, and poverty. AI regulations should therefore include means of making sure that these systems do not have the opposite of their intended effect.

The dynamics of climate change and AI

AI technologies need to be developed in such a way as to mitigate their climate risks; especially since they are likely to be used increasingly to address the effects of climate change in the future.

In a few months, the UNESCO ethics framework will be published, and possibly lead to the development of an international convention on AI. The outcomes of the COP26 might serve to further strengthen the environmental considerations in this standard-setting document for AI and all future regulatory instruments.

Mis à jour le 20 novembre 2021 à 14 h 36 min.


Laisser un commentaire